Back to vendors
I

Imbue

Visit site
Agent builderindependentVerified 2026-07-01

Desktop app from the AI research lab Imbue for running parallel coding agents in isolated containers, focused on verifying the correctness of the code those agents generate.

Imbue is a San Francisco AI research lab, formerly known as Generally Intelligent and founded in 2018, that builds tools to make AI coding reliable, collaborative, and accessible. Led by chief executive Kanjun Qiu, it has raised roughly two hundred thirty two million dollars from investors including NVIDIA, Eric Schmidt, Cruise founder Kyle Vogt, and Notion co founder Simon Last, and it operates a ten thousand graphics processor research cluster. Its research mission is to improve the ability to verify the correctness of code generated by large language models, so people can trust the software they build.

Imbue's first product is Sculptor, a desktop application for Mac and Linux that provides the missing interface for running parallel coding agents. Each agent runs in its own isolated container, so multiple agents can execute code, install dependencies, and run tests safely in parallel without contaminating a developer's machine or colliding with each other. Sculptor runs coding agents like Claude Code and OpenAI Codex, letting developers choose the right model for the job and bring their own key. It is designed to let a developer kick off a refactor with one agent while another builds a feature and a third checks a hunch.

Its most distinctive features center on control and verification. Pairing Mode syncs any agent's work from its container into the local editor with one click, so changes can be tested and reviewed instantly, and merges flag conflicts for the agent to resolve. Instruction audits let a developer write rules in plain English, like never use a particular function, and Sculptor reviews the agent's work to catch violations, while its Suggestions feature flags issues such as tests that falsely report passing. Every session is saved with its plans, chats, tool calls, and code changes intact, so context carries across sessions.

Sculptor runs fully locally, with granular control over what data returns to Imbue and no training on a user's code by default. It is open source and available in beta, free while in beta with a bring your own model key arrangement, though long term pricing is not yet published. As a focused developer tool rather than a general business agent platform, it does not offer enterprise security certifications, broad enterprise integrations, channel coverage, or an agent marketplace. Its strength is orchestrating and, above all, verifying coding agents, which is precisely the problem Imbue set out to solve.

Vendor details

Canonical URL

https://imbue.com

Category

Agent builder

Company status

independent

Use cases & customers

In practice

A developer wants to explore three approaches at once. In Sculptor they spin up three agents, each in its own container, kicking off a refactor, a new feature, and a bug fix in parallel, then jump between them without environment conflicts.

An engineer worries an agent faked a passing test. Sculptor's Suggestions flag the misleading behavior, and an instruction audit rule written in plain English catches other violations before the code is merged into the local repository.

A developer returns to a project after a week. They reopen the saved Sculptor session with its plans, chats, tool calls, and code changes intact, and pick up exactly where the agent left off without re prompting.

Sources & related URLs

Research notes

AI research lab (San Francisco), formerly 'Generally Intelligent', founded 2018. CEO Kanjun Qiu, co-founder Josh Albrecht. Raised $232M (NVIDIA, Eric Schmidt, Kyle Vogt/Cruise, Simon Last/Notion). 10,000 H100 GPU cluster. Research mission: VERIFY correctness of LLM-generated code. FIRST PRODUCT = Sculptor: desktop app (Mac/Linux, Windows-WSL) for running PARALLEL coding agents in isolated Docker containers. Runs Claude Code + OpenAI Codex/GPT-5 (BYOK: bring your own Anthropic/OpenAI key or Claude Pro/Max plan). Key features: Pairing Mode (bidirectional sync agent-container↔local IDE, test instantly), sandboxed containers (safe parallel exec/deps/tests), session persistence (save plans/chats/tool-calls/code-changes, resume across sessions, fork agents), Instruction audits (write plain-English rules e.g. 'don't use eval', Sculptor reviews agent work to catch violations), Suggestions (flag misleading behavior like 'tests passed' without real tests), integrates linters (pylint/ruff/pyre). Runs fully LOCALLY, levels (Essential/Standard/Full), no-train-on-code by default. Open source (github.com/imbue-ai/sculptor). SCORING: Orch=F (parallel coding-agent orchestration), Eval=F (CORE thesis — verify LLM-code correctness, run tests until pass, audit rules, flag fake-passing-tests). Mem=P (session-state persistence + resume + fork — rare). Sec=N (local dev tool, no enterprise certs/SSO/RBAC/governance). Pack=N (few workflow/repo templates, no marketplace). Trig=N (manual on-demand only). Comp=N (code/terminal execution ≠ browser/GUI computer-use). NARROW developer coding-agent tool, NOT a general business agent-builder — FLAG category fit for Mike. Beta/research-preview, free+BYOK. Domain imbue.com. Score 6.0 (2F/8P/4N).

Capability coverage

6.0 / 14 capabilities · 43%

Integrations & Tool CallingSculptor connects to code repositories, editors, and common linters and frameworks and runs agents with terminal and code execution, real developer tool integration short of broad enterprise system connectivity, so partial. Partial
Workflow OrchestrationSculptor runs multiple coding agents in parallel in isolated containers, delegating tasks and coordinating their work end to end from building features to fixing bugs and merging changes, so full. Full
Knowledge Grounding & RAGSculptor's coding agents are grounded in a connected code repository and its context, real grounding in the codebase short of a documented citation grounded knowledge retrieval system, so partial. Partial
Human Oversight & GuardrailsSculptor requires human review of every change before applying and enforces plain English instruction audit rules that review agent output for violations, strong oversight short of a hard runtime blocking enforcement engine, so partial. Partial
Security, Identity & GovernanceSculptor runs locally in isolated containers with granular data sharing controls, but named enterprise security certifications, identity governance, and role based access could not be verified. Unable to verify
Observability & AuditabilitySculptor saves every agent session with its plans, chats, tool calls, and code changes intact and makes agent contributions inspectable, real observability short of a documented production monitoring and audit system, so partial. Partial
Memory & State PersistenceSculptor saves each agent session's plans, chats, and context and lets users resume across sessions and fork from any point, real cross session state persistence short of a documented shared long term agent memory, so partial. Partial
Deployment & Data ResidencySculptor runs fully locally on a user's own machine with granular control over what data is shared back, real local deployment and data control short of a documented enterprise on premises or dedicated residency offering, so partial. Partial
Prebuilt Agents, Templates & PacksSculptor offers a few prebuilt coding workflows and repository templates, but a browsable library or marketplace of prebuilt agents could not be verified. Unable to verify
Triggers & Channel CoverageSculptor launches coding agents on demand from a desktop application, but scheduled or event driven triggers and broad channel coverage could not be verified. Unable to verify
Model Flexibility & RoutingSculptor supports Claude Code and OpenAI Codex with the ability to choose the right model for the job and bring your own key, multi provider model selection short of a documented routing gateway, so partial. Partial
APIs, SDKs & MCP ExtensibilitySculptor is open source and lets users define custom checks with shell commands and plain English rules, a real extensibility surface, though a public software development kit, application programming interface, or Model Context Protocol support is on the roadmap rather than shipped, so partial. Partial
Testing, Debugging & OptimizationSculptor is built around verifying the correctness of agent generated code, running tests in sandboxes until they pass, auditing agent work against plain English rules, and flagging issues like tests that falsely report passing, a genuine testing and verification framework, so full. Full
Browser & Computer UseSculptor agents execute code and run terminals in isolated containers, but general browser or graphical computer use as a first class capability could not be verified. Unable to verify

Recent platform changes

No recent material changes tracked yet.

Pricing

Free in beta (bring your own model key)

Public — exactMedium variable cost
Verified 2026-07-01

Contact us

Found a vendor we missed? Have feedback on the index? We'd love to hear from you.

Agentic AI Index

A directory and comparison resource for AI agent platforms, autonomous workflow tools, and enterprise agentic automation products.

© 2026 Agentic AI Index

3801 N Capital of Texas Hwy, Ste E240 · Austin, TX 78746

Researched from public vendor sources. See Methodology.